Multi-agent deep reinforcement learning based multi-timescale voltage control for distribution system

• Main Author: Wang, Bingyu
• Other Authors: Soong Boon Hee
• Format: Thesis-Master by Research
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Electrical and electronic engineering
• Online Access: https://hdl.handle.net/10356/159271
• Institution: Nanyang Technological University

As low-carbon and clean energy become an inevitable requirement for sustainable development of energy, modern distribution networks are integrating more and more renewable energy resources, mainly in the form of rooftop solar photovoltaics (PV) panels. As a DC generation source, the solar PV is interfaced with the grid through power electronics inverters. Apart from converting DC power to AC power, the PV inverters can also generate and absorb reactive power for voltage/var control (VVC) purposes. In this work, a data-driven multi-timescale volt-var control (VVC) framework has been proposed to counteract uncertain voltage fluctuation and deviation caused by PV energy integration. An MDP model has been built to describe the multi-timescale voltage control problem. A multi-agent deep deterministic policy gradient (MADDPG) method has been used to solve the model. Compared with the conventional VVC method, the proposed method has a faster response speed and a better result. The proposed method is verified on the IEEE 33-bus distribution network and compared with existing practices. In this work, the author uses python to run the multi-agent deep reinforcement learning program. And let python uses the MATPOWER toolbox in Matlab. This result is also compared with multi-agent DQN learning to see the outstanding of this proposed method.